An ontology-based metamodel for multiagent-based simulations

https://doi.org/10.1016/j.simpat.2013.09.002Get rights and content

Abstract

Multiagent-based simulations enable us to validate different use-case scenarios in a lot of application domains. The idea is to develop a realistic virtual environment to test particular domain-specific procedures. This paper presents our general framework for interactive multiagent-based simulations in virtual environments. The major contribution of this paper is the integration of the notion of ontology as a core element to the design process of a behavioral simulation. The proposed metamodel describes the concepts of a multiagent simulation using situated agents moving in a semantically enriched 3D environment. The agents perceive the geometric and semantic data in the surrounding environment. They are also able to act in this environment by using high-level actions, which are described by the ontology of the environment. The concepts relating to the environment, the agent, and the entire simulation models are presented. Additionally, guidelines are given to exploit the simulation results to characterize the agents. Finally, a simple application of the metamodel is presented, based upon the use of Industry Foundation Classes.

Introduction

Multiagent-based simulations are now a common tool in various application domains to assess/validate different use-case scenarios: social sciences, urban management, transport, building security assessment, etc. The idea is to develop a realistic virtual environment to test particular domain-specific procedures. Developing realistic scenarios implies that they must be non-deterministic and agents inhabiting virtual environments are autonomous and intelligent. The variety of their behavior must reflect the heterogeneity of human behavior.

This paper presents our general framework for interactive multiagent-based simulations in a virtual environment (2D or 3D). Its main objective is to integrate the notion of ontology as a core element of the design process of a behavioral simulation in order to facilitate its use/reuse by simulation designers and end-users in many application fields. To ease the use of a multi-agent simulation, it should be easily configurable: each agent behavior must be at high level and application-independent, and have their configurations as automated as possible. For empirical studies, multiagent-based simulation results are generally integrated and compared with results from other types of simulation (discrete events, finite elements, etc.) to merge the different views available on the studied system. It is therefore interesting to have a common representation of the simulation results that is both interoperable and cross-domain.

We consider that the key point to address these issues is to have a semantic description of the environment in a multiagent-based simulation (MABS). To do this, this paper presents a new approach coupling multi-agent systems and semantic modeling with ontology.

Ontology is a general term for a semantic modeling of knowledge to define the know-how. Using ontologies, the system can infer new knowledge and relations from existing resources.

To enable the development of high-level easy to configure agent behavior, it is important to provide agents with the means to reason on their surronding environment. The agents must be able to analyze unexpected situations to dynamically adapt their behavior to achieve their personal goals [5]. Semantic rules and the agent’s reasoning procedures can enable the development of such smart behavior.

For example, if you plan to send an agent to the restaurant, just specify that your agent is hungry. With semantic rules, the agent decides that he must eat. This action is semantically linked with “restaurant” (place where agents can eat in the ontology) and then he will go, in the environment, in a place defined as a restaurant. This solution saves a lot of designing and configuration time by just modeling the agent as hungry instead of designing a whole behavioral plan. The plan is dynamically determined at runtime according to a succession of semantic rules.

The designing of the agents behaviors is thus simplified thanks to the usage of semantic. But to enable this kind of reasoning behavior, the environment must provide agents with a semantic description of themselves.

Reasoning mechanisms and dynamic adaptation procedures of agents behavior are out of the scope of this paper. In this article, we describe the essential components that must provide the environment of a multiagent-based simulation to allow agents to easily support these mechanisms. This paper presents a new metamodel for MABS exploiting situated agents evolving in a semantically enriched 3D environment. This metamodel can be considered as a generic ontology [20].

The proposed approach is illustrated on a simple example derived from typical use cases in the field of building qualification (which is one of our main targeted application areas) [3], [1], [4]. In this example, environment description is based on IFC (Industry Foundation Classes) files, which are a standard in the building industry.

They describe buildings in both a geometrical and semantic manner. This semantic description allows to easily build the semantic structure of the simulation environment.

The essence of our proposal is an ontology describing the semantic of every element needed or produced during a simulation (agents, environment, interactions, etc.) This ontology integrates the common MABS standards like the influences/reactions model [21], [16], [17], [8], the clear separation between an agent’s mind and body, etc. This paper presents a formal modeling of this semantic metamodel that includes the main simulation principles, how to represent agents, manage their interactions, etc.

The paper is organized as follows: after a quick introduction to your ontology-based and agent-oriented metamodel (Section 2), the different components of the proposed metamodel are successively detailed (3 Environment, 6 Results analysis: characterization of the agents). It is then illustrated on a simple scenario in Section 7. Section 8 describes related works. Finally, Section 9 summarises the contributions of the paper and describes some future work directions.

Section snippets

A quick overview of the ontology-based metamodel

The goal of our metamodel is to integrate every key element required for developing multiagent-based simulations. It is now widely recognized that a MABS may be split into at least four main parts [14], [16], [17]:

  • Agent behavior: modeling of the agent’s deliberative process.

  • Environment: definition of the various physical objects composing the simulated world as well as the endogenous dynamics of the environment. It is here that a number of fundamental principles must be respected to guarantee a

Environment

The term “Environment” defines both the physical environment and the physical description of agents (agent’s body). Four main missions are classically assigned to the environment in a mabs [14], [21]:

  • Sharing information. The environment is first of all a shared structure for agents, where each of them perceives and acts. This shared structure can also be the support of indirect communication (i.e. stigmergy).

  • Managing agents’ actions and interactions. For this aspect, the action model in

Agents

There are many agent definitions in the literature. These approaches differ on the characterization of agents for the scope of the simulation.

We identify common concepts from these heterogeneous approaches.

An agent is an autonomous entity that progresses in an environment and able to interact with it. The various definitions mainly differ on three fundamental features: autonomy, behavior and interaction. The definitions of the environment are, for their part, a dedicated research topic and are

Simulation

This section presents the principles of the simulation, its life-cycle, according to our metamodel. The gathering of the simulation logs is also explained.

Results analysis: characterization of the agents

The metamodel allows to apply some existing agent definitions and to retrieve agents that correspond to a given definition during the simulation. This characterization is made according to the characteristics that are bound to the agents. The simple association of a characteristic is not enough to perform the characterization when the characteristic is about an interaction. Indeed, the ability for an agent to perform an interaction is involved in the characterization process, not only the

Application

This section describes a simple application of the proposed metamodel. This application takes place in the field of a situated multiagent-based simulation. The agents are situated in a 3D building. For the building representation, the Industry Foundation Classes (IFC) norm is used.

The next section describes the IFC format and usage. The scene used during the illustration is then illustrated and the goal is explained. After that, the environment, and then the agents, are described. The run is

Background

This section compares our metamodel with other models using ontologies for multiagent-based simulations. It is difficult to find proposals which combine a representation of agent, environment and interaction dedicated to a large range of multiagent-based simulations. Existing models are usually domain-dependent or focused on a specific part of the simulation.

In 1997, Franklin and Graesser [12] listed many agent-related approaches and definitions and tried to establish a taxonomy of agents.

We

Conclusion

In this paper, we have presented a new ontology-based metamodel for agent-based simulation that fully exploits the advantages of ontologies.

This model fits the standards of multiagent-based simulations by respecting the principle of influence/reactions [8], [17], the separation between an agent’s body and mind [9], [13], the principle of behavioral planning based on ontologies [5], etc. In short, it ensures a clear separation of concerns, between the different parts of the simulation in order

Acknowledgements

This work is supported and funded by the Regional Council of Franche-Comté, France. We thank Caroline Tabard for her help in reviewing this paper.

References (21)

There are more references available in the full text version of this article.

Cited by (0)

1

http://www.multiagent.fr.

2

http://www.checksem.fr.

View full text